Target product profiles

Ganga Rao Nadigatla, Harish Gandhi

Genetic gain: On-farm yield estimation (RTB)

One of the goals of on-farm testing is to get insights into genetic gain achieved by breeding programmes. Some aspects of genetic gain are related to traits that are highly heritable so that on-farm performance is not different from on-station performance. For example, the color of the product may not be affected by genotype by environment interactions. An aspect of genetic gain that is important as a goal shared by most breeding programmes is the yield. As tricot is based mainly on rankings, generally yield estimations have been provided in that form. This provides an insight into the yield-based reliability, the probability that a new variety will outperform the current market leader, an important indicator for breeders and product managers to make decisions (Eskridge and Mumm, 1992). The CGIAR Excellence in Breeding strategy focuses on product profiles that emphasize cumulative gains towards product replacement, taking over market share from existing varieties (Cobb et al., 2019). Tricot is well suited to address the challenge of providing early indicators of the probability that product replacement happens.

In many cases, however, breeders need to have absolute estimates of yield levels, for example because this is a requirement for a variety release procedure. In one case, a subset of the fields has been visited to obtain yield estimates (NextGen Cassava), in other cases, all fields were visited for yield measurements (de Sousa et al., 2020). This ‘undermines’ the tricot approach to some degree in the sense that the field visits become an important cost driver. This leads to the question whether farmers themselves can provide reliable yield estimates.

Ochieng, Ojime, and Otieno (2019) have addressed this question by comparing yield estimates by researchers (taking into account grain moisture) and by farmers (volumetric, using 250 ml tins). They set up an experiment with common bean (P. vulgaris) in Kenya. They obtained a high correlation between the two types of measurements when all seasons and locations were aggregated (r = 0.98). When differences were smaller than 0.5 t/ha, the match between values provided by farmers and researchers decreased. We aim to replicate these studies in other contexts with other crops in order to get a better grip on the accuracy of farmers’ measurements and to use these accuracy estimates in statistical analyses. On the other hand, these studies will provide insights in how to maximize farmers’ accuracy. It would be ideal to be able to combine yield ranking data and yield measurement data when the measurement data is only available for a part of the trial. It is possible to feed absolute measurements and ordinal (ranking) data into the same statistical model, directly (Böckenholt 2004) or through a Bayesian approach. This has not been implemented in software yet; this is a pending task.